Thread: Filesystem setup on new system

Hi list,


I'm helping a customer with their new postgresql server and have some
questions.

The servers is connected to a SAN with dual raid cards which all have
512MB cache with BBU.

The configuration they set up is now.
2 SAS 15K drives in RAID 1 on the internal controller for OS.

6 SAS 15K drives in RAID 10 on one of the SAN controllers for database

10 SATA 10K drives in RAID 5 on the second SAN controller for file
storage.

My first idea was to have one partition on the RAID 10 using ext3 with
data=writeback, noatime as mount options.

But I wonder if I should have 2 partitions on the RAID 10 one for the
PGDATA dir using ext3 and one partition for XLOGS using ext2.

Should I do this instead? Is there just a minor speed bump as the
partitions are on the same channel or?

If this is the way to go how should it be configured?

How big should the xlog partition be?

How should the fstab look like for the XLOG partition?

Any other pointers, hints would be appreciated.

Cheers,
Henke

On Thu, 7 Aug 2008, Henrik wrote:

> My first idea was to have one partition on the RAID 10 using ext3 with
> data=writeback, noatime as mount options.
>
> But I wonder if I should have 2 partitions on the RAID 10 one for the PGDATA
> dir using ext3 and one partition for XLOGS using ext2.

Really depends on your write volume.  The write cache on your controller
will keep having a separate xlog disk from being as important as it is
without one.  If your write volume is really high though, it may still be
a bottleneck, and you may discover your app runs better with a dedicated
ext2 xlog disk instead.

The simple version is:

WAL write volume extremely high->dedicated xlog can be better

WAL volume low->more disks for the database array better even if that
mixes the WAL on there as well

If you want a true answer for which is better, you have to measure your
application running on this hardware.

> 6 SAS 15K drives in RAID 10 on one of the SAN controllers for database

With only 6 disks available, in general you won't be able to reach the WAL
as a bottleneck before being limited by seeks on the remaining 4 database
disks, so you might as well group all 6 together.  It's possible your
particular application might prefer it the other way though, if you're
doing a while lot of small writes for example.  I've seen a separate WAL
handle low-level benchmarks better, but on more real-world loads it's
harder to run into that situation.

--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD